The DJT Solver represents a breakthrough in the extremely difficult problem of transportation optimization. The DJT engine can solve very large scheduling problems in near real-time (within seconds).
The engine was originally developed to support an innovative new air carrier: DayJet Corporation. The scientific team worked for 5 years on developing algorithms to support an air-taxi model (i.e. no fixed flight schedule) that required solving a large combinatorial problem in seconds. During development the team quickly realized that the algorithms were applicable to a broader range of problems in transportation and other complex assignment and configuration problems.
The design uses a multi-grid approach that is superior in both speed and quality to the existing standard methods. Using this engine we can build applications that manage a large number of assets in real-time. Fleets can be managed optimally while demand changes and disruptions occur.
Most optimization systems used in transportation are based on linear- or mixed-integer programming. Programs such as CPLEX are used in batch fashion and require significant computation for non-trivial problems (i.e. greater than 50 orders). Applications using these programs do not easily adapt to changes that occur in the course of operation.
The DJT engine scales very well to real-world problems. It has been tested on problems with over 1500 vehicles and 15,000 requests. Solutions are computed within seconds on today’s hardware. This capability is significantly better than currently available systems.
The algorithms are designed to take advantage of the latest parallelization and virtualization technologies. Search subspaces and hypotheses can be partitioned and explored on different processors or virtual machines leading to deeper optimization within seconds.
The DJT algorithms and architecture allow for both deep optimization and rapid recovery from changes:
- The solver maintains one or more feasible solutions.
- A new request arrives:
- New demand (reservation, shipment) or a
- Disruption (change order, cancelation, equipment breakdown)
- The solver repairs the solution/s.
- Batch processes can be used to further fine tune optimization.